In its 2½ years of public life artificial intelligence (AI) has evolved at such a speed that it’s hard to get the head around how it might change the lives of its human hosts, for better or worse. But we do know this: it is upsetting our best-laid plans for a future energy supply.
Intelligent or not, AI clearly has an exceptional ability to scan the internet for what it can offer and then, all within seconds, receive and process queries and put results into an easily understood, cohesive form. Impressive to say the least. And before you ask, no – AI didn’t write this.
To achieve this feat it needs large, sustained supplies of electrical energy – so much that it has upended estimates of future demand. AI is transforming global energy markets in ways few envisaged before ChatGPT was unleashed on the public in November 2022.
AI’s rapid expansion is forcing fundamental rethinking in every corner of public policy including corporate regulation, the stock market, the labour market, productivity, creative arts and copyright, criminal activity and national security. But the biggest shock so far has been felt by energy grid managers and those charged with devising plans and targets for lowering carbon emissions.
“Power hungry”, a Massachusetts Institute of Technology study published in May, plots the exponential rise of AI energy demands. It points out that early on, as AI models learned what data was available, the focus was on their training. But as AI became a global phenomenon it moved to a new phase, inference – its response to the queries of you and me. That’s where most power is used.
In the words of an MIT explainer, “the emissions from individual AI text, image, and video queries seem small – until you add up what the industry isn’t tracking and consider where it’s heading next.” It added that as AI technologies become more personalised and ubiquitous their energy demands will only grow, making it likely that “our AI footprint today is the smallest it will ever be”.
Spending on data centres being planned or built – trillions of dollars of private and public capital – is currently propping up an otherwise flagging US economy. This is hard to fathom, but many individual centres draw as much energy as a large city. Globally, both AI power demand and AI capital spending are currently around or above the level of an entire economy like Australia’s.
This is a big departure from electricity use in the recent past, when increasing energy efficiency enabled data centres to expand without enlarging the overall energy footprint. Right now, the big US companies driving the global AI push are drawing heavily on carbon-intensive fossil fuel energy because that’s what’s available.
But looking ahead, those companies are realising they are going to need an unprecedented amount of new energy, starting now, deployed quickly and at lowest cost. Therein lies a whole universe of trouble for energy supply and planning. Only the solar-battery option can meet those needs. Creating new fossil fuel plants takes too long, and nuclear much longer still.
Entrepreneurs heading many of these companies snuggled up to Donald Trump ahead of last year’s presidential election, polishing his image by helping him to look tech-savvy and even largely bankrolling his campaign.
It’s hard to imagine this egocentric bunch admitting such a thing, but they’ve been well and truly hoodwinked. Trump’s administration is hard at work turning America into a fully-fledged petrostate, busily pulling funding from the best source of AI energy – solar and battery research and manufacture.
China’s path to the present has been somewhat different. We don’t know how advanced its AI program is, but the country’s decades-long, economy-wide focus on renewables and batteries, ideal for powering AI, is now paying dividends, making solar cheaper for the whole world. But Trump’s tariffs, by design, are making Chinese products more expensive in that country. Oops.
Like everywhere else, energy demand here is soaring. The Australian Energy Market Operator reported a new demand record for 2024, breaking a mark set in 2008. This has been driven mainly by federal Labor’s electrification program based on Climate Change Authority advice to accelerate renewable energy rollout.
AI’s inference phase is one of several factors in this demand surge, but with more data centres it will soon become the dominant one, requiring a step increase in electricity generation. Australia’s focus on renewables, if their continuity of supply issues can be overcome, puts us in an excellent position to become a global leader in meeting the AI challenge.
As for emissions targets, let’s just say they’re a moving feast.